I'm currently using this page's code to build image captioning model. The only different is I'm using BERT model's output as word embedding at Decoder part, therefore I have to change some of the original parameter like d_model, dff and target_vocab_size. The training went fine but when I tried to use evaluate function, the caption contained repeated words. Any help would be really appreciated. Below is my Decoder:

class Decoder(tf.keras.layers.Layer):

def __init__(self, num_layers, d_model, num_heads, dff, target_vocab_size,
               maximum_position_encoding, rate=0.1):
   super(Decoder, self).__init__()

   self.d_model = d_model
   self.num_layers = num_layers
   self.embedding = TFBertModel.from_pretrained('bert-base-uncased')
   self.pos_encoding = positional_encoding_1d(maximum_position_encoding, d_model)
   self.dec_layers = [DecoderLayer(d_model, num_heads, dff, rate) 
                   for _ in range(num_layers)]
   self.dropout = tf.keras.layers.Dropout(rate)

 def call(self, x, enc_output, training, look_ahead_mask=None, padding_mask=None):
   seq_len = tf.shape(x)[1]
   attention_weights = {}
   bert_output = self.embedding(x)  # (batch_size, target_seq_len, d_model)
   x=bert_output[0]

  x *= tf.math.sqrt(tf.cast(self.d_model, tf.float32))
  x += self.pos_encoding[:, :seq_len, :]

  x = self.dropout(x, training=training)
  for i in range(self.num_layers):
     x, block1, block2 = self.dec_layers[i](x, enc_output, training,
                                         look_ahead_mask, padding_mask)

  attention_weights['decoder_layer{}_block1'.format(i+1)] = block1
  attention_weights['decoder_layer{}_block2'.format(i+1)] = block2
 return x, attention_weights
       

Here are my parameter:

num_layer = 12
d_model = 768
dff = 9216
num_heads = 12
row_size = 8
col_size = 8
target_vocab_size = 29690
dropout_rate = 0.1